The AI startup has provided a more granular breakdown of its infrastructure costs in a revised registration statement for its initial public offering, disclosing a minimum spend of $330B on compute purchase and lease agreements through 2030. The updated S-1 filing reveals that $437B of the company’s aggregate commitments for AI hardware and data center access are non-cancellable, highlighting the rigid nature of the spending required to scale its large language models.
The financial disclosures coincide with new reporting on the company’s technical targets. While Anthropic has recorded nearly $900B in total compute obligations, the New York Times reports the company now expects to reach more than 3.5 GW of computing capacity by the end of 2026 and approximately 7 GW by 2027. These targets represent a roughly 30% decrease from the capacity goals the company shared with investors in September.
